Everything About AI
NextFin News - Goldman Sachs upgraded CoreWeave to buy on Monday and lifted its price target to $139 from $106, the clearest signal yet that at least one major Wall Street firm views last week's credit-market rout as a misread of the AI capital-spending cycle rather than a warning about demand. The upgrade arrived as the artificial-intelligence complex staged a tentative rebound: the Nasdaq Composite climbed 1.11% to 26,431 and the S&P 500 added 0.42% to 7,691, led by chips and AI-linked names after the worst weekly selloff in months. Nvidia rose 1.38% to $217.09, and neocloud peers Nebius Group and IREN closed up 2.88% and 2.74% respectively.
CoreWeave itself was the axis of the storm. The neocloud provider's stock fell 25% last week, from $82 on Friday, October 9, to $61 on Friday, October 16, with a 15% single-day drop on Wednesday as its bonds slid toward distressed levels. By Monday's close the shares were at $61.12, down 0.44% on the day but off more than 50% from their May highs and 30% since September 30. Trading volume reached roughly 100.2 million shares, about 198% above the three-month average of 33.3 million - the footprint of a market that is still arguing with itself. The question the upgrade forces investors to answer is whether the bond market priced a company on the brink, or whether it priced a panic that the equity market is now trying to unprice.
What Goldman Actually Argued
The upgrade is a bet on a spread. In a Monday note, Goldman Sachs analyst Gabriela Borges framed CoreWeave's valuation as a lesson in "Finance 101": the gap between what the company can charge for its chip-rental business and what it costs CoreWeave to borrow money. Her call rests on the view that investors are "underestimating the magnitude of upward revisions" to the return the company earns on its hardware while "overestimating the magnitude of upward revisions" to its cost of capital.
The mechanism is straightforward, and it runs in both directions. A shortage of processing power should lift the revenue CoreWeave can earn per megawatt of data-center capacity over the next 24 months, which is why Goldman sees average recurring revenue per megawatt as biased to the upside. On the liability side, Borges estimates that even new debt will come at only 100 to 200 basis points above current levels, capping CoreWeave's blended borrowing cost at around 10%. "Our conversations with credit specialists suggest that CoreWeave should still be able to finance its future expansion plans at a cost of debt that is only modestly higher than its current weighted average cost of debt," Borges wrote.
The structure of the balance sheet matters to that argument. CoreWeave's delayed-draw term loans sit inside separate corporate vehicles, collateralized by physical hardware and guaranteed by customer contracts from investment-grade names such as Meta Platforms and Microsoft. That debt is designed to amortize over the life of each contract, which means the lenders' first claim is on contracted cash flows from creditworthy tenants rather than on CoreWeave's corporate credit. The unsecured notes that triggered the distress alarm represent less than $2 billion of a roughly $35 billion total debt stack, according to Barclays analyst Ajay Rajadhyaksha, who drew a direct parallel to the March 2023 regional-banking panic.
"Even if the bonds might be rationally priced for their own recovery prospects they could still be being completely wrong as a signal about the viability of the entire AI capex cycle," Rajadhyaksha wrote. "As long as that's true, the financing follows the demand."
That is the crux of the bull case: the bond market was pricing the recovery value of a small slice of unsecured paper, while the equity market extrapolated that signal into a verdict on the whole AI infrastructure buildout. Dan Ives of Yorkville Ives made the same point last week, describing the company's roughly $104 billion revenue backlog - reported in August - as commitments customers have already made, and calling the drawdown a buying opportunity.
Why the Selloff Happened, and Why It Was Rational
The counter-thesis deserves the weight the market gave it. CoreWeave's five-year credit-default swaps hit a record 1,194.5 basis points last week, and its 2032 notes traded near 79.5 cents on the dollar. The yield on its unsecured debt reached 15%, a level that typically signals distress. Those are not hallucinations; they are prices at which lenders were willing to transact. A CDS print above 1,000 basis points implies the market is pricing a substantial probability of default within five years, and a bond trading below 80 cents is, by convention, in distressed territory. The equity market ignored those signals at its peril.
The underlying arithmetic is also uncomfortable. Total debt reached $51.61 billion at June 30, including about $16.5 billion of lease obligations. Second-quarter interest expense hit $640 million, up roughly 140% from a year earlier, against a $49 million operating loss. Management has guided third-quarter interest expense to between $860 million and $940 million. Interest, in other words, is growing faster than the business is proving it can earn. The market's fear is not that demand for AI compute is fake; it is that CoreWeave's cost of capital may outrun the spread Borges is betting on.
There is precedent for this kind of credit-led repricing turning into something worse. When a capital-intensive business funds growth with debt and the debt market closes, the growth stops - and a backlog is only as valuable as the financing available to build it. CoreWeave reported a record $104 billion backlog in August, but a backlog of commitments does not generate cash until the capacity exists to deliver it. If tighter credit prevents the next round of data-center construction, the backlog can evaporate or migrate to competitors with cheaper balance sheets: the hyperscalers themselves, or neocloud rivals still able to borrow.
This is where the cyclical-versus-structural call determines the conclusion. The selloff was cyclical in its trigger - a spike in yields after hotter-than-expected inflation data, transmitted through a thin, leveraged balance sheet - and cyclical moves revert when the trigger fades. But the leverage is structural. CoreWeave chose a debt-funded buildout in an industry where the cost of capital is the decisive competitive advantage, and that choice does not reverse. The right read is therefore split: the panic was cyclical and likely overdone, but the constraint it exposed is structural and will not disappear on its own.
The Second-Order Trade the Market Has Not Fully Priced
The first-order effect of the rebound is obvious: oversold AI stocks bounce, shorts cover, and headlines about a $139 price target steady sentiment. The second-order effect is more consequential. If the market accepts Goldman's framing - that the credit signal was noise and the demand signal is intact - then the dispersion within AI infrastructure should widen, not narrow. Capital will migrate toward the names with the cheapest financing and the strongest counterparties, and away from the leveraged fringe. The AI buildout may continue, but it will not reward every participant equally.
Nvidia's chief executive, Jensen Huang, tried to make exactly that point on Thursday, saying he was optimistic about CoreWeave and that its cost of capital was actually decreasing. A supplier with a 10% equity stake in the customer has an incentive to say so, but the incentive also aligns: Nvidia needs CoreWeave to keep buying GPUs, and a CoreWeave that cannot finance its buildout is a Nvidia customer that stops ordering. Retail investors appear to agree with the bounce thesis; data from Robinhood Markets showed record five-day net purchase volume in the stock during the decline.
Yet the rebound itself was imperfect, which is the tell. CoreWeave opened strong, rose as much as 7.4% intraday, then gave up its gains and closed fractionally lower at $61.12. A clean reversal on an upgrade from a top-tier bank would have held the gains. Instead the tape showed two markets arguing: the upgrade buyers in the morning, the credit-conscious sellers into the close. That split is the market's way of saying the debt question is not settled by a price target, however prestigious the source.
The peer read supports the dispersion thesis. Nebius Group closed up 2.88% at $170.24 and IREN added 2.74% to $27.73, while Nvidia rose 1.38% to $217.09. The complex is recovering, but not uniformly - and the names that recover fastest will be those whose financing story is cleanest. This is the second-order consequence that a simple "AI stocks bounce" headline misses: the rebound is not a rising tide for all boats. It is a sorting event.
The neocloud model explains why the sorting matters. A neocloud operator rents out GPU capacity on medium-term contracts, often to hyperscalers that need compute beyond what they can build themselves. The economics are a levered bet on utilization: fixed hardware costs funded with debt, variable revenue priced by scarcity. When GPUs are scarce, per-megawatt revenue rises and the leverage works in the operator's favor; when scarcity eases or financing costs rise, the same leverage works against it. CoreWeave's second-quarter gross margin of roughly 24% leaves room for interest costs to consume a large share of the upside, which is why every 100 basis points on the debt side matters as much as every percentage point on the revenue side.
What Would Prove the Upgrade Wrong
The strongest case against the Goldman call attacks it at the foundation: the spread could compress from both sides at once. If AI demand softens - if the hyperscalers slow their capital expenditures because their own returns disappoint - then revenue per megawatt falls just as refinancing costs rise. In that scenario the "Finance 101" spread that underpins the upgrade narrows, and a 10% blended borrowing cost stops looking manageable. This is the same dynamic that has made leveraged infrastructure businesses fragile across cycles, from pipelines to telecom towers: the asset is real, the demand is real, and the math still does not work when the cost of capital moves against you.
The falsifying signal is specific. Watch CoreWeave's next quarterly interest expense against its revenue growth, and the pricing of its unsecured notes. If interest expense prints at the top of the $860 million to $940 million guidance range while revenue per megawatt fails to rise, or if the 2032 notes trade back below 80 cents on the dollar, the credit market is telling you the equity rebound is a bear-market rally rather than a bottom. A second consecutive week of five-year CDS above 1,000 basis points would confirm that the distress signal was fundamental, not noise. Conversely, if the notes recover above 90 cents and CDS falls back toward 700 basis points, the bond market will have signaled that the panic has passed.
What Comes Next
Short term, the rebound is a sentiment and liquidity move: oversold positioning unwinds, retail steps in, and an upgrade from a top-tier bank provides cover. That can carry shares higher for days even if nothing fundamental has changed. The stock is trading near its 52-week low of $59.25, and technical buyers often defend such levels on the first bounce.
Medium term, the story belongs to the earnings report due in late October. Investors will be looking for three things: revenue per megawatt, the trajectory of interest expense, and any update on the $104 billion backlog and its conversion into deployed capacity. The base case is that CoreWeave reports strong demand and confirms that financing remains available at manageable cost, which would validate Goldman's call and extend the rebound. The downside case is that interest costs consume more of the margin than expected, forcing the company toward dilutive equity issuance or slower buildout. Analysts surveyed place the average 12-month price target near $140, so the Goldman upgrade moves the stock toward consensus rather than dramatically above it.
The wider AI capital-expenditure backdrop sets the stage for that report. The hyperscalers - Alphabet, Microsoft, Meta and Amazon - have collectively committed well over $200 billion a year to AI infrastructure, and their own commentary has framed the spending as a race for scarce capacity rather than a discretionary budget line. That is the demand floor the bulls are relying on. But the same commentary has also begun to emphasize returns, and any shift from "build at any cost" to "build where returns clear" would hit the leveraged operators first.
Long term, the structural question dominates: in an industry where the winner is the operator with the lowest cost of capital, a balance sheet that funds growth at 15% marginal rates is a competitive disadvantage no matter how strong the backlog. The AI buildout is likely to continue; which companies capture it is the open question. Hyperscalers with investment-grade balance sheets can fund their own capacity more cheaply over time, which is why the neocloud model only works if the processing shortage persists long enough to justify the premium.
The market rebounded on Monday because it decided the bond market had overreacted. The safer conclusion is narrower: the bond market overreacted about demand, but it was right about leverage - and those are two different verdicts that will not resolve on the same timetable.
Explore more exclusive insights at nextfin.ai.
